An Ontology Design Pattern to Define Explanations
Ilaria Tiddi, Mathieu d’Aquin, Enrico Motta
Knowledge Media Institute, The Open University, United Kingdom
{ilaria.tiddi, mathieu.daquin, enrico.motta}@open.ac.uk
ABSTRACT
In this paper, we propose an ontology design pattern for the
concept of “explanation”. The motivation behind this work
comes from our research, which focuses on automatically
identifying explanations for data patterns. If we want to
produce explanations from data agnostically from the application domain, we first need a formal definition of what an
explanation is, i.e. which are its components, their roles or
their interactions. We analysed and surveyed works from the
disciplines grouped under the name of Cognitive Sciences,
with the aim of identifying differences and commonalities in
the way their researchers intend the concept of explanation.
We then produced not only an ontology design pattern to
model it, but also the instantiations of this in each of the
analysed disciplines. Besides those contributions, the paper
presents how the proposed ontology design pattern can be
used to analyse the validity of the explanations produced by
our, and other, frameworks.
Categories and Subject Descriptors
I.2.4 [Artificial Intelligence]: Knowledge Representation
Formalisms and Methods — Ontology Design Patterns
Keywords
Explanation, Ontology Design Pattern, Knowledge Discovery
1.
INTRODUCTION
“The word explanation occurs so continually and holds so
important a place in philosophy, that a little time spent in
fixing the meaning of it will be profitably employed.”
John Stuart Mill – A System of Logic, 1843.
In this paper, we present an ontology design pattern to
support the formal representation of an explanation. The
motivation behind our work comes from our research, whose
aim is to automatically find an explanation to data patterns using background knowledge from the Web of Data.
Note that a data pattern is intended as a subset of data
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for components of this work owned by others than the
author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from [email protected].
K-CAP 2015 October 07 - 10, 2015, Palisades, NY, USA
c 2015 Copyright held by the owner/author(s). Publication rights licensed to ACM.
ISBN 978-1-4503-3849-3/15/10. . . $15.00
DOI: 10.1145/2815833.2815844
behaving in the same regular way. Our general research
problem is that finding an explanation for those patterns is
an intensive task which is still relying on human experts,
whose role is to provide explanations and refine results using their own background knowledge, while the cross-domain
and machine-accessible knowledge of the Web of Data could
be used to facilitate the automatisation of explaining data
patterns.
The challenge we face here is that if we wish to give explanations to patterns or, more specifically, to process data in
order to produce them, we first need a formal way to define
what is an explanation E 1 . Defining E is a complicated epistemic matter upon which a common agreement has never
been reached, although this has been extensively discussed
through time and across disciplines. Rather than entering
this discussion, our proposition is to identify an ontology
design pattern that formally supports the representation of
an explanation – including its components, their roles and
interactions – in order to provide an abstract description
which can be applicable to any context where a system automatically produces explanations.
The methodology we used to derive such a design pattern
is to explore how explanations are defined in those areas
that most deal with the organisation and understanding of
knowledge, usually grouped and defined as the “cognitive
sciences”. We reviewed the main literature in the disciplines
embraced by Cognitive Science in order to see how their researchers see explanations. By identifying differences and/or
commonalities among different areas, such as which aspects
of an explanation matter and how E is defined, we aimed at
abstracting our own definition for E.
What we finally propose is a formal definition of E encoded as an ontology design pattern (odp), presented with
its instantiations into ontological representations of explanations in each of the considered disciplines of Cognitive
Science. Those constitute, along with a cross-domain survey on the definition of explanation, the major contributions
of our paper. Finally, we also show how to use the pattern
to assess the validity of a framework producing explanations
in real-world scenarios.
2.
MOTIVATION
In our previous work [37], we presented Dedalo as a framework to explain groups of items using background knowledge
extracted from Linked Data2 . Dedalo’s assumption is that
1
We will refer to E further on, to intend “the concept of
explanation”.
2
http://www.w3.org/standards/semanticweb/data
items appearing in the same group share underlying commonalities beside the dimensions used for their grouping,
and these commonalities can be detected in the graph of
Linked Data in the form of chains of RDF properties and
a value shared by a large part of the items in the group
to explain. We call those chains “candidate explanations”.
Because Linked Data can be traversed using URI dereferencing, Dedalo iteratively expands the graph using an A*
strategy trying to find new, more accurate chains (evaluated with an ad-hoc F-Measure). Finally, a Linked Data
pathfinding approach is used to establish the relation between the extracted candidate explanations and the criteria
that grouped the items, in order to identify the most plausible explanations while removing noisy results.
In a very short example, we might want to understand why
people search for the actor Daniel Radcliffe only at specific
times3 . The explanation, if being familiar with the context, can be identified as “people search for Daniel Radcliffe
when a Harry Potter movie is released”. While a human
uses his own background knowledge to derive this (therefore, explaining the phenomenon), Dedalo uses the one from
Linked Data. The problem here is that, if the human is not
familiar with the context of Harry Potter, he will not find
the above statement as a plausible explanation. Statistically
computing explanations on the Linked Data graph has also
its drawbacks, especially because some coincidences could
emerge in the set of candidate explanations (e.g., “people
search for Daniel Radcliffe when the U-20 Football World
Cup is on”4 ).
From the example above, we understand that an explanation does not only mean putting two events in a correlation
(“X happened because of Y”), but also defining the causal
dependence based on the context in which they occur (“X
happened because of Y in some context C”). The question is
then whether this is enough to define an explanation and, if
not, what else is needed for E to be complete. In a broader
sense, in this work we aim at identifying which are the important actors of an explanation, i.e. which elements and
interactions are needed to let us declare that, given a pattern and the derived observation, we have raised them to
an explanatory level, not only to the one of a pure correlation. Our challenge consists then in designing an ontology
pattern to formally represent it. In order to do so, we explored how the concept of explanation is defined within the
disciplines embraced in Cognitive Science (in Section 3) to
find their commonalities and differences in the view of an
explanation and finally extract and apply (in Section 4) an
ontology design pattern to define E.
3.
DEFINING EXPLANATIONS: A SURVEY
Cognitive Science was firstly defined by [16] as the science of understanding the nature of the human mind. In
this work, dated 1985, Cognitive Science is presented as a
field involving a series of disciplines interacting in such a
way that they can be represented as a “cognitive hexagon”
(Figure 1). The next sections provide a summary of how
the main trends in these disciplines have tried to define explanations. It is worth mentioning that this is an organi3
See the search peaks shown by the Google Trends tool, at
http://www.google.com/trends/explore#q=Daniel%20Radcliffe
4
The release of some Harry Potter movies and more than
one U-20 football World Cup have indeed happened at the
same time.
Figure 1: The cognitive hexagon as in [16]. Dotted
lines are weaker connections between disciplines.
sational choice of our methodology, but many of the cited
works, especially the contemporary ones, span over multiple
disciplines. Additionally, we chose to consider the broader
and more common fields of Computer Science and Sociology
rather than the original Artificial Intelligence and Anthropology. The common nomenclature, that we also adopt, is
to define explanandum/nda (“that which is explained”) the
fact or event to be explained and explanans/tia (“that which
does the explaining”) some premise to the explanandum.
3.1
Explanation and Philosophy
The works of [28, 30, 36] provide interesting surveys on
how explanations have been defined by intellectuals from the
ancient times – the first attempts of discussion already appear among the Greek intellectuals – to contemporary days.
Thucydides (in his History of the Peloponnesian War ) defines explanations as a process where facts, or “indisputable
data”, are observed, evaluated based on our knowledge of human nature and then compared in order to reach generalised
principles for why some events occur. For Plato (in Phedus
and Theaetetus), an explanation is an expression of knowledge using the logos, and is composed by the Forms, consisting in the abstractions of the entities we know (“the unchanging and unseen world”) and the Facts, i.e. occurrences
or states of affairs, which are the changing world we are used
to see. Aristotle in his Posterior Analytics presents the explanation as a deduction process of finding out the cause of
why something happened (according to Aristotle’s 4-cases
doctrine, the cause can be either the thing’s matter, form,
end, or change-initiator). In the modern age of determinism,
causality becomes prominent to explanations. To know what
caused an event means to know why it happened, which
in turns means understanding the universe determined by
natural laws. Descartes and Leibniz describe explanations
as the process of demonstrating the mechanical interactions
between God (“the primary efficient cause of all things”),
the world things (secondary causes), the laws of nature and
some initial conditions. Newton rejects the God component
and puts the laws of nature as playing the central role. For
him, explanation of natural phenomena is a deductive process which finds the most general principles (the fundamen-
tal laws of nature) that account for them. The Empiricists
then reject explanations which cannot be traced directly to
experience. According to Hume, all causal knowledge stems
from experience, and causation is just a regular succession
of facts to be discovered by experience. Kant, reintroducing the metaphysical component into explanations, believes
that the knowledge starts with experience, but does not arise
from it: It is shaped by the categories of the understanding
and the forms of pure intuition (space and time). John Stuart Mill rejects this view and defines E as only based on
the laws of nature and on deduction. A fact is explained
by deducing its cause, that is, by stating the laws of which
it is an instance. Carl Hempel, who revisited and extended
Mill’s work, started the contemporary discussions around
explanations. Explaining an event is to show how this event
would have been expected to happen, taking into account
the laws that govern its occurrence, as well as certain initial conditions. Formally speaking, a singular explanandum
E is explained if and only if a description of E is the conclusion of a valid deductive argument, whose explanantia
involve essentially a set C of initial or antecedent conditions
and a law-like statement L [20]. L can be either statistical
(in inductive-statistical explanations) or laws of nature (in
deductive-nomological explanations). Salmon [32] further
extends the inductive-statistical explanation model proposing the statistical-relevance model, where explanations have
to capture the dependencies between the explanandum and
the explanans statistically, and the causal model, which sees
explanations as mechanical processes of interacting components. An interactive view is also the one of the Interventionists [39], where explanations are the process of manipulating the explanandum within a space of alternative possibilities to see if a relation holds when various other conditions change. The Unificationism approach of Friedman
and Kitcher [13, 23] affirms that explaining a phenomenon
is to see it as an instance of a broad pattern of similar phenomena. According to Friedman, explanations are based on
both the scientific understanding of the world, but also on
some assumptions (i.e. regularities in nature) which are the
basic “unifiers”. Similarly, Kitcher believes that a number
of explanatory patterns (or schemata) need to be unified in
order to explain a fact.
3.2
Explanation and Psychology
Explanations in psychology are generally meant to regulate behaviours or to process some information, i.e. they
are “an attempt to understand phenomena related to intelligent behaviours” [2]. Many works of psychology criticise
the traditional philosophical-scientific approaches to explanation since they are too much focused on causality and
subsumption under laws (deduction) [2, 8, 21]. Instead,
psychological explanations accept over-determinism and dynamism, i.e. the antecedent conditions that explain a phenomenon can be more than one and change over time. The
explanantia are not laws of nature but human capacities,
that psychological methodologies have to discover and confirm. In the literature, psychological explanations have been
defined as “models” providing a law, by means of which it is
possible to describe a phenomenon while relating it to other,
similar phenomena, and possibly to allow a prediction, intended as a future occurrence of the phenomenon [8, 21].
Authors of [26] have highlighted how the mechanical aspect
within the explanation process plays a central role into the
recent psychology works. Phenomena are often explained by
decomposing them into operations localised in parts (components) of the mechanism. Mechanisms have spatial and
temporal organisations that explain how entities are organised into levels and carry out their activities. In other words,
explaining a phenomenon means revealing its internal structure by defining the organisational levels which are responsible of producing it, then identifying how those levels relate to
other levels, and finally building a model to explain similar
phenomena. This idea of interactive levels can be found in
the major contemporary psychology trends, i.e. the connectionist (bottom-up) approach [2, 9, 25], that first specifies
the system’s components and then produces the explanation’s model, and in the symbolic (top-down) one [34], that
first identifies the phenomenon to be explained and then
describes it according to the external symbols (syntacticosemantic rules or mathematical laws) manipulated by the
humans.
3.3
Explanation and Neuroscience
Neuroscience explanations reject the concept of levels, as
well as the synthetic a priori frameworks. Mentality is the
central pivot between interpretation and rationality. Neuroscientists aim at understanding minds, and this requires to
assume that beliefs are true and rational, which is in contrast with many philosophical views. Explaining, in neuroscience, means fitting a mental phenomenon into a broadly
rational pattern. Explanations are answers to the whatif-things-had-been-different questions: they describe what
will happen to some variables (effects) when one manipulates or intervenes on some others (causes) [4, 10, 40]. They
can be “upper level explanations”, when causes are macroscopic variables or environmental factors, and “lower level
explanations”, if causes are more fine-grained (neural, genetic, or biochemical) mechanisms. Also, explanations do
not require any law nor sufficient condition, and are simply generalisations that describe relationships stable under
some appropriate range of interventions. Therefore, a typical approach to explanations consists in detecting how the
variables respond to the hypothetical experiments in which
some interventions occur, then establishing if other variables
under the same intervention respond in the same way. The
relevance of explanations is assessed by their stability, i.e.
some relationships are more stable than others under some
interventions or under some background conditions.
3.4
Explanation and Computer Science
In Artificial Intelligence (AI), explanations are considered
in terms of inference and reasoning. Programs are meant
to decipher connections between events, in order to predict
other events and represent some order in the universe. The
process of generating explanations fits the two reasoning patterns, abduction and induction, that Peirce describes in its
Lectures on Pragmatism of 1903. According to Peirce, induction is the inference of a rule (major premise) starting
from a case (minor premise) and its result (conclusion), while
abduction is the inference of the case from a rule and a result [12]. Those models have been often put under the same
umbrella-term of non-deductive, a posteriori inference, and
the same Peirce in its later period sees abduction as only
one phase of the process, in which hypotheses and ideas
are generated, but then need to be evaluated by induction
or deduction [12, 31, 33]. Moving on from such discus-
sion, AI approaches to generate explanations (see [27, 29,
31] for extensive surveys) show that the two inferences can
be similarly modelled, so that one body of initial knowledge
(divided into observations and prior knowledge) is mapped
to another (the new knowledge), under constraints of certain criteria, which reveal their connection and confirm the
hypothesis. The explanation process then consists in first
analysing the observations, then defining tentative hypotheses and finally proving them empirically. Similarly to explanations in psychology, Artificial Intelligence explanations
do not assume that a truth has been antecedently accepted
in the prior knowledge: In order to produce new knowledge, they need to be scientifically and empirically evaluated. Plausibility, usefulness and desirability are considered
too subjective and are replaced by some formal constraints
that the process has to respect, namely, efficiency (how “easily” it solves the problem), non-deducibility (it should introduce new knowledge), consistency (it must be consistent
with what it is already known) and non-emptiness (it must
introduce new knowledge). Some works attempting to merge
the views from philosophy and Artificial Intelligence are surveyed in [27]. Explanations also play a central role in other
areas of Computer Science such as Knowledge Management.
Knowledge Management systems are used for intelligent tutoring or decision-support, and explanations provide insight
into how their reasoning is accomplished. Users are presented solutions and explanations of a problem to be solved
relying on some domain knowledge (see referenced works
in [1, 35, 41]). This domain knowledge, related to facts, procedures or events, is defined as “tacit” because it is already
in the individuals’ mind. An inflow of new stimuli then triggers a cognitive process which produces explanations, which
are the explicit knowledge communicated in symbolic forms.
Explanation can be intended as “giving something a meaning
or an interpretation to, or to make it understandable” [41].
Linguistic explanations are descriptions focused on finding what rules of a language explain a linguistic fact [19].
Linguists do not aim at describing the processes but rather
at explaining how language acquisition is possible, by identifying the grammatical patterns that occur in human languages. Useful surveys on explanations in linguistics are to
be found in [7, 18, 19, 22]. A central assumption in linguistic
theories is that a language is a set of elements (words, morphemes, phonemes) that meet a set of wellformedness conditions, called language universals. Those consist in languagespecific characteristics that form the grammar of the language. The major approaches, the generative linguistics
(following Chomsky’s theories) and the functional linguistics (based on Greenberg’s school) have different views on
the roles and functions of the components of linguistic explanations. In the former case, only the generalisation of an
explanation is regarded as interesting, since it shows that
linguistic constructions are derivable from the general regularities of a language (descriptive adequacy), but some of
those regularities are innate and have to be accepted a priori (explanatory adequacy). The generativists see the act
of explaining as defining new constraints on the frameworks
which are able to describe the languages. This claim to
innateness and the view of explanations as constrained descriptions cannot be accepted by functional linguistics, for
which external elements and realities are essential to understand linguistic facts. Functionalists are more interested in
language regularities than in descriptive frameworks. The
rejection of a priori language universals does not mean that
in language structure is arbitrary, but rather than it is worth
attempting to explain, wherever possible, language facts using external non-linguistic factors [18].
3.5
3.7
Explanation and Sociology
For Weber and Durkheim [11, 38], founders of the major two trends in sociology, explaining is the act of giving a
meaning and justifying social facts, where those are intended
as observable regularities in the behaviour of the members
of a society. In their perspective, explanation is a rational
(empirical) observation of social behaviours. More contemporary works extend this view, by defining explanations as
a comparison of events [5] or as a form of reasoning [3],
which have to take in account the social world and social
behaviours constantly evolving in time and space. Generally, the agreement is that explanations, based on social
facts, have more pragmatic requirements with respect to
scientific explanations. Formal constraints, such as mathematical formalisation or empirical falsifiability, leave space
to descriptions and approximations of complex structures of
the social world [6, 17]. Explanations might be weaker or
stronger depending on the set of social impacts that the rules
express. For example, they can be proverbs (everyday rules
expressing what one knows), social theorems (rules summing
up the human experience) or paradoxes [3, 24]. Some have
also distinguished between social-anthropological and religious explanations, where “religious” means giving sense to
phenomena using symbols of a tradition. Explanations are
declared to be valid if a phenomenon can also appear due
to other circumstances, and logically exhaustive if they are
generalisable to all logically possible combinations of condi-
tions encountered [3, 24].
3.6
Explanation and Linguistics
Summary Considerations
From the sections above, we briefly summarise the following conclusions.
• Philosophy is the “discipline of asking questions and checking answers” [16]. An explanation in Philosophy is a process where a set of initial elements (an event and some
initial conditions) are described, deduced or put into a relation with an output phenomenon according to a set of
(empirical or metaphysical) laws.
• Psychology tries to understand the human (and animal)
cognitive processes. Here, E is intended as a model of components (entities and activities) that mechanically interact
to produce regular changes from some starting conditions
to a termination condition.
• Neuroscience gives an understanding of the brain. Explanations are seen as generalisations that describe relationships between input and output variables, that are stable under some appropriate range of interventions without
any law nor sufficient condition.
• Computer Science deals with computer programs as a
medium to perform human operations. Explanations are
generalisations of some starting knowledge (some facts
and some prior knowledge) mapped to another, new knowledge under constraints of certain criteria.
• Sociology focuses on the links between the human’s cognitive processes and his sociocultural world. It intends
Figure 2: The Explanation Ontology Design Pattern.
explanations as the act of giving a meaning to some social
regularities occurring because of some social behaviours.
• Linguistics provides theories dealing with the nature of the
human language. An explanation is the process of deriving
the regularities of a language (the grammar) from a set
of linguistic-specific facts respecting some wellformedness
constraints.
With those premises, it is possible to abstract a common
model for E. The linguistic formalisation we found particularly suitable for our purposes was the one of [24] and presented hereafter:
When X happens, then, due to a given set of circumstances
C, Y will occur because of a given law L.
The next section is focused on the design of the ontology
pattern representing E, as well as its instantiations in the
various presented disciplines.
4.
THE EXPLANATION DESIGN PATTERN
Ontology Design Patterns are small, motivated ontologies
to be exploited as building blocks in ontology design [15].
Their purpose is to be used as designing components of a
problem with the assumption that an ontology should be
modelled according to those patterns, with appropriate dependencies between them, plus some necessary design expansion based on the specific situation’s needs. We followed
this idea to build an ontology design pattern that represents
an explanation. We reused some of the existing design patterns and ontologies, and then extended classes and properties according to our needs. The resulting model, that we
call E-odp, is shown in Figure 2 and described hereafter5 .
4.1
Ontology Design Pattern
Shortly, we define E-odp as an ontology for characterising
what is an explanation to an event, and include the various
components that make the process plausible (for instance,
the initial conditions under which the explanation happens,
the outcome event, and so forth).
(a) The main class Explanation represents E, which we aim
to identify. It will be differently instantiated according to the disciplines we have presented. The class has
few constraints that we represent as OWL class restrictions. In order to be complete, an explanation: (i)
needs at least one antecedent event, expressed as hasExplanans some part:Event; (ii) requires a posterior event,
5
Many thanks to Silvio Peroni (University of Bologna) for
its help in designing the presented pattern.
expressed as hasExplanandum some part:Event and (iii)
has to happen in a context that relates the two events,
expressed as hasCondition some situation:Situation.
(b) An explanation has then an antecedent event (the explanans) and a posterior event (the explanandum). We
saw how those have been called not only events but also
phenomena, facts, variables, observations, knowledge,
etc. We chose to reuse the part:Event class from the
Participation odp6 , which enables to represent any binary relation between objects and events. Thanks to
this, our pattern can be easily integrated in the Participation one by extending the part:Event individuals
according to the structure of the Participation odp.
(c) Two events have to be related in context, otherwise it is
not possible to distinguish their correlation from a coincidence. To represent the conditions (or constraints)
under which the two events occur, we used the situation:Situation class from the Situation odp7 , which is
precisely designed to “represent contexts or situations,
and the things that have something in common, or are
associated” (see documentation). The OWL axiom in
the blue box above the object property hasCondition is
a Manchester Syntax axiom showing what we can infer
about the situation in which the explanation is happening. If there exists a situation in which an explanation
is contextualised (through hasCondition), then both the
explanans and the explanandum do share this same situation (through situation:hasSetting), therefore they are
in the same context.
(d) The last main component of an explanation is represented using the dul:Theory class from the DOLCE+DnS
Ultralite ontology8 . According to the documentation, “a
theory is a description that represents a set of assumptions for describing something, usually general. Scientific, philosophical, and common-sense theories can be
included here”. We use the term theory with the idea
that it best embraces those concepts defined in Cognitive Science (laws, human capacities, human experiences, universals, and so on), all consisting in something having a binding force on some events under analysis. Note that the Theory is a specialisation of the
Description class from the Descriptions&Situations ontology [14]: In this view, our theory acts as the description that classifies the explanation (corresponding to the
6
http://ontologydesignpatterns.org/cp/owl/participation.owl
http://www.ontologydesignpatterns.org/cp/owl/situation.owl
8
http://www.ontologydesignpatterns.org/ont/dul/DUL.owl
7
Figure 3: Instances of the pattern in Philosophy (left) and Psychology (right).
situation), which is built upon the context, the event to
be explained, and the possible explaining events.
(e) We also include the DOLCE class dul:Agent to represent
the one producing the explanations. Since there is no
agreement on how to “call” the act of explaining (e.g.,
deducing, inducing, inferring, describing have all been
used in the literature), we then kept the generic object
property label as dul:isConceptualizedBy.
4.2
Instances of the Pattern
Finally, we create instances of the E-odp according to
our findings in the survey of Section 3. We present the
ontological model for each discipline as well as a practical example which gives an idea of how the process works.
In the examples, we use the notation E=(A, P, C,T ) where
A stands for the antecedent event/explanans, P for the
posterior event/explanandum, C for the situational context
they are happening in and T for the theory governing those
events. For the purpose of avoiding redundancy, the class
dul:Agent is omitted. Note that all the presented patterns
have been submitted to the ODP central hub9 but are also
available online with a higher resolution10 .
Philosophy. The pattern for E as seen in Philosophy is
presented in Figure 3 (left). T is a Law, that we also subspecified as metaphysical or empirical, trying to sketch some
of the general views in the area (roughly, pre- and postempirical views). Both A and P are Phenomenon classes
while C is a necessary and sufficient Condition class. It is
worth mentioning that ours is a mere generalisation of how
an explanation is thought in Philosophy, and that this can
be in turn specialised into more specific patterns according
to, for instance, different times or trends. For example, we
could represent Mill’s explanation model as Mill E = (CausingFact, EffectFact, InitialCondition, LawOfNature) while
Plato’s model as Plato E = (InitialFact, PosteriorFact, Circumstance, Form).
Psychology. The pattern for E as intended in Psychology
is presented in Figure 3 (right). We have shown in Section
3 how psychological explanations are more focused on their
mechanical aspects. Therefore, A is an InitialComponent
class, whose subclasses are the interacting classes Entity and
Activity. The context C is the human IntelligentBehaviour
that correlates A and P. The theory consists in a HumanCapacity and not in laws of nature. In a practical example,
a simple event P as “slipping on a banana” can be explained
by the fact A that the person did not to pay attention to the
path. A psychological explanation would attempt to relate
9
10
http://ontologydesignpatterns.org/
http://people.kmi.open.ac.uk/ilaria/experiments/explain odp/
the two events to (C) the mental state and feelings of the
person slipping as, for instance, anger or tiredness. If this
condition is not met anymore (e.g., one is calm and careful),
the explanation does not hold. The human capacities which
govern the examples are the ones ruling the human’s mind
so that the human actions change depending on the feelings.
The remaining patterns are shown in Figure 4 in a clockwise
order.
Neuroscience. The idea of explanations in Neuroscience is
similar to the one in Psychology, with the difference that the
events are considered as CauseVariable (A) and EffectVariable (P). We have seen how neuroscientists reach explanations by Experiment(s) (T ) rather than with laws, and that
the necessary conditions C are rather called Intervention(s)
under which the two events occur. If those are changed,
they might nullify the explanation. For example, we explain
why humans can do math calculations (P) by the fact A
that some particular neurons respond actively to quantities.
The neuroimaging has proven that (T ) some areas on the
parietal lobe are used to provide pre-verbal representations.
The context C in which the explanation holds is the one of
numerical calculations.
Computer Science. The model for an explanation in
Computer Science can be easily represented as CSE =(Observation, NewKnoweldge, Constraint, PriorKnowledge). Note
that the new knowledge can be GeneralisedNewKnowledge,
if the explanation is reached by induction, or SpecialisedNewKnowledge, if reached by abduction. We use here the
example given by [31]: a caveman is roasting a lizard (C) on
the end of a pointed stick and he is watched by an amazed
crowd of cavemen, who have been using for years only their
bare hands. The cavemen explain that painless roasting (P)
is effect of (A) using a long, rigid, sharp object simply by
observing (which is equivalent to using the prior knowledge
T ) how the stick supports the lizard while keeping the hand
away from the fire.
Sociology. Sociology reintroduces the human component
into its idea of explanation. We represent it as SE =(SocialBehaviour, SocialRegularity, SocialWorld, HumanExperience). The social trend P that Italians live with their parents until later ages is explained by (A) the constantly decreasing job opportunities, which makes impossible living on
its own (C). The theory behind it are the rules governing
the socio-economical behaviours.
Linguistics. Explanations in Linguistics are models focusing on the function and nature of the language, namely a
series of LinguisticFact(s) specific to a Language are analysed on the basis of some LanguageUniversal (s) to finally
Figure 4: Instances of the pattern for (clockwise) Neuroscience, Computer Science, Sociology and Linguistics.
derive a Grammar. For example, to explain that English
(C) does not allow the expression “*the my book ” (P), one
has to know that both my and the are determiners (A), and
that (T ) only one determiner is accepted.
4.3
Pattern Application on Real Frameworks
Once we have derived the pattern for E, we can return
to our original motivating scenario and analyse how much
Dedalo’s framework fits into the process of deriving an explanation. The same approach can be of course applied to
any other framework that automatically produces explanations. The scenario, already presented in Section 2, is the
following: Dedalo wants to explain the fact P that a group
of users is searching the Web for information about Daniel
Radcliffe only at some very specific times. What we need to
identify is what constitutes A, T and C.
The first component of Dedalo uses the Linked Data background knowledge to derive some plausible explanantia of
type A to the event P. The process consists in finding
facts in Linked Data that are statistically highly correlated
with the explanandum. For example, both “people search for
Daniel Radcliffe because a Harry Potter movie is released”
and “people search for Daniel Radcliffe because the Under20 Football World Cup is on” are correlated with Daniel
Radcliffe according to Linked Data information11 . If no connection between the A events and P is identified, however,
there is no way to distinguish whether the A and P are happening together by coincidence (as in the second case). Also,
the explanation is incomplete (and impossible) if one has no
knowledge of the context C (what is the relation between
Daniel Radcliffe and Harry Potter).
Dedalo’s second component is therefore the one finding
this relationship, in the form of a Linked Data path between
two entities representing the facts (for instance, Daniel Radcliffe and Harry Potter/the U-20 Football World Cup) to
11
To test this scenario, see http://linkedu.eu/dedalo/demo
provide a more explicit information to the user. For example, there exists a Linked Data path such as hDanielRadclif
fe portrays HarryPotteri but not between Daniel Radcliffe and the U-20 Football World Cup. This path explicits
the context C in which those two facts are happening: without it, the two events are nothing but a spurious correlation.
This brings us to our conclusion point. The obtained explanation can be summarised as “people search for Daniel
Radcliffe because a Harry Potter movie is released, since
Daniel Radcliffe is the main character in it”. A familiar
reader might recognise here the law T governing the explanation: actors play in movies, therefore it makes sense
that people are interested in Daniel Radcliffe if he is the one
playing in Harry Potter movies. Dedalo is, however not (yet)
able to provide the same information: the user to whom the
explanation is presented is left inferring that Daniel Radcliffe
is an actor, that “Harry Potter” is a movie and, equally, that
“actors play in movies” (as a general principle).
5.
CONCLUSIONS
This work is an attempt to formally define what is intended as an “explanation”. We have presented such formalisation as an ontology design pattern, whose modelling
has been achieved by analysing how the different disciplines
of Cognitive Science perceive explanations, i.e. which elements and interactions are needed for an explanation to be
in place. The proposed ontology pattern has then been instantiated according to the surveyed disciplines. Finally, we
have shown how to apply the pattern on custom frameworks
that automatically produce explanations in real-world scenarios, by showing how to assess whether the system fits the
proposed pattern.
We identified as an important future work the introduction in our Linked Data-based framework of a third component, whose role is to explicit the laws underlying the
relationship between the events, so that the explanation pre-
sented to the users can be considered complete according to
the proposed E-odp. Another direction to explore is the possibility of refining the pattern by introducing which qualities
are necessary to an explanation (e.g., validity, truth, and so
on) or by exploring new perspectives (e.g., other disciplines).
Finally, an evaluation of other existing frameworks on the
basis of our ontology design pattern is also possible.
6.
REFERENCES
[1] M. Alavi and D. E. Leidner. Knowledge management
systems: issues, challenges, and benefits.
Communications of the AIS, 1(2es):1, 1999.
[2] W. Bechtel and C. Wright. What is psychological
explanation. Routledge companion to the philosophy of
psychology, pages 113–130, 2009.
[3] E. Brent, A. Thompson, and W. Vale. Sociology a
computational approach to sociological explanations.
Social Science Computer Review, 18(2):223–235, 2000.
[4] J. Campbell. Causation in psychiatry. Philosophical
issues in psychiatry, pages 196–215, 2008.
[5] J. R. Carter. Description is not explanation: A
methodology of comparison. Method & Theory in the
Study of Religion, 10(2):133–148, 1998.
[6] P. Clayton. Explanation from physics to the
philosophy of religion. International journal for
philosophy of religion, 26(2):89–108, 1989.
[7] P. W. Culicover and R. Jackendoff. Simpler syntax.
Oxford University Press Oxford, 2005.
[8] R. Cummins. How does it work?” versus” what are the
laws?”: Two conceptions of psychological explanation.
Explanation and cognition, pages 117–144, 2000.
[9] L. Darden and N. Maull. Interfield theories.
Philosophy of Science, pages 43–64, 1977.
[10] D. Davidson. The structure and content of truth. The
journal of philosophy, pages 279–328, 1990.
[11] E. Durkheim. The rules ofthe sociological method,
1938.
[12] K. T. Fann. Peirce’s theory of abduction. Springer
Science & Business Media, 1970.
[13] M. Friedman. Explanation and scientific
understanding. The Journal of Philosophy, pages 5–19,
1974.
[14] A. Gangemi and P. Mika. Understanding the semantic
web through descriptions and situations. In On The
Move to Meaningful Internet Systems 2003: CoopIS,
DOA, and ODBASE, pages 689–706. Springer, 2003.
[15] A. Gangemi and V. Presutti. Ontology design
patterns. In Handbook on ontologies, pages 221–243.
Springer, 2009.
[16] H. Gardner. The mind’s new science: A history of the
cognitive revolution. Basic books, 2008.
[17] P. S. Gorski. The poverty of deductivism: A
constructive realist model of sociological explanation.
Sociological Methodology, 34(1):1–33, 2004.
[18] M. Haspelmath, H.-J. Bibiko, J. Hagen, and
C. Schmidt. The world atlas of language structures,
volume 1. Oxford University Press Oxford, 2005.
[19] J. A. Hawkins. Explaining language universals.
Blackwell, 1988.
[20] C. G. Hempel and P. Oppenheim. Studies in the logic
of explanation. Philosophy of science, pages 135–175,
1948.
[21] E. H. Hutten. On explanation in psychology and in
physics. The British Journal for the Philosophy of
Science, 7(25):73–85, 1956.
[22] E. Itkonen. On explanation in linguistics. In article in
the forum discussion of this issue: http://www.
kabatek. de/energeia/E5discussion, 2013.
[23] P. Kitcher. Explanatory unification. Philosophy of
Science, pages 507–531, 1981.
[24] E. Maaløe. Modes of Explanation: From Rules to
Emergence. Handelshøjskolen, Aarhus Universitet,
Institut for Ledelse, 2007.
[25] D. Marr and A. Vision. A computational investigation
into the human representation and processing of visual
information. WH San Francisco: Freeman and
Company, 1982.
[26] M. Marraffa and A. Paternoster. Functions, levels, and
mechanisms: Explanation in cognitive science and its
problems. Theory & Psychology, page
0959354312451958, 2012.
[27] A. Páez. Artificial explanations: the epistemological
interpretation of explanation in ai. Synthese,
170(1):131–146, 2009.
[28] S. Psillos. Past and contemporary perspectives on
explanation. General philosophy of science. Focal
issues, pages 97–173, 2007.
[29] N. J. Rothleder. Induction: Inference and process,
1996.
[30] D.-H. Ruben. Explaining explanation. Cambridge Univ
Press, 1990.
[31] S. Russell, P. Norvig, and A. Intelligence. Artificial
intelligence. a modern approach. Artificial Intelligence.
Prentice-Hall, Egnlewood Cliffs, 25, 1995.
[32] W. C. Salmon. Scientific explanation: causation and
unification. Critica: Revista Hispanoamericana de
Filosofia, pages 3–23, 1990.
[33] G. Schurz. Patterns of abduction. Synthese,
164(2):201–234, 2008.
[34] P. Smolensky. On the proper treatment of
connectionism. Behavioral and brain sciences,
11(01):1–23, 1988.
[35] R. W. Southwick. Explaining reasoning: an overview
of explanation in knowledge-based systems. The
knowledge engineering review, 6(01):1–19, 1991.
[36] M. Strevens. Scientific explanation. Encyclopedia of
Philosophy, second edition. DM Borchert (ed.).
Detroit: Macmillan Reference USA, 2006.
[37] I. Tiddi, M. d’Aquin, and E. Motta. Dedalo: Looking
for clusters explanations in a labyrinth of linked data.
In The Semantic Web: Trends and Challenges, pages
333–348. Springer, 2014.
[38] M. Weber and W. V. Heydebrand. Sociological
writings. Continuum International Publishing Group
Ltd., 1994.
[39] J. Woodward. Making things happen: A theory of
causal explanation. Oxford University Press, 2003.
[40] J. Woodward. Cause and explanation in psychiatry.
Philosophical issues in psychiatry. Explanation,
phenomenology, and nosology, pages 132–184, 2008.
[41] B. A. Wooley. Explanation component of software
system. Crossroads, 5(1):24–28, 1998.